INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Smart Pest Monitoring and Management System with Integrated  
Deep Learning and Unmanned Aerial Vehicle (UAV) Technologies  
Ogidi Patient C., Asogwa T.C  
Department of computer science, Enugu State University of Science and Technology, Nigeria  
DOI: https://dx.doi.org/10.51584/IJRIAS.2025.10100000182  
Received: 02 November 2025; Accepted: 10 November 2025; Published: 21 November 2025  
ABSTRACT  
Infestation of pest is one of the leading agricultural problems which have led to a lot of losses in the yield and  
risks food security. The use of conventional forms of pest control is usually inefficient, consumes a lot of  
chemicals and requires response time. The paper describes a smart pest monitoring and management system,  
combining the deep learning, Internet of Things (IoT) and Unmanned Aerial Vehicle (UAV) technologies to  
detect pests in real-time and give decision support. The annotated mixed data consisting of primary pest images  
gathered in the Federal College of Agriculture, Ishiagu, and secondary data in the Kaggle repository was used to  
generate a model based on YOLOv10. The model had a precision of 0.84, a recall of 0.82 and a mean Average  
Precision (mAP@50) of 0.83 indicating that the model was very effective in detecting and classifying a variety  
of pest species at an acceptable level of accuracy. A recommendation algorithm based on rules was installed to  
offer specific pesticide recommendations depending on the identified type of pest and the IoT-based email  
notification module provided the real-time notifications to the farmers to take immediate action. To realise  
remote sensing and aerial pest surveillance, the UAV was simulated and designed in the Simulink environment  
to ensure the efficient coverage and the reliable data capture. The integrated system offers a smart and long-term  
solution to the pest management process by eliminating false alarms, reducing pesticide waste, and enhancing  
the reaction time. The limitation of the study lies on the condition that the system was only being implemented  
as a simulation and lacks real-world validation, hence, it is recommended that future studies should adopt the  
real-world implementation approach for the identification of pests.  
Keywords: Smart Agriculture; Pest Detection; YOLOv10; Internet of Things (IoT); Unmanned Aerial Vehicle  
(UAV); Deep Learning  
INTRODUCTION  
Pests in agriculture are organisms that harm crops, livestock, or agricultural infrastructure, posing a threat to  
food production and economic stability. These pests include a wide range of species such as insects, rodents,  
fungi, weeds, bacteria, and viruses, all of which can cause significant damage to crops by reducing yield, quality,  
and overall productivity (Karar et al., 2022). Insect pests, for example, feed on plant tissues, while fungal  
infections can cause disease that rots crops. As a result, pest management has become a crucial aspect of modern  
farming practices aimed at safeguarding agricultural output (Masood, 2023). The impact of pests on agriculture  
extends beyond crop damage; they also affect food security and environmental sustainability. For instance, pest  
infestations can lead to severe crop losses, which ultimately drive-up food prices and affect both local and global  
food supply chains (Alanazi et al., 2023). Additionally, pests can reduce the quality of harvested products,  
making them unsuitable for consumption or sale, leading to economic losses for farmers.  
Managing pests in agriculture requires a comprehensive approach that encompasses traditional methods such as  
chemical pesticides and biological controls as well as modern technologies. Integrated pest management (IPM)  
aims to use a combination of strategies to control pest populations while minimizing environmental harm (Anwar  
and Masood, 2023). However, growing resistance to chemical pesticides and the adverse effects of excessive  
chemical use on the environment have pushed the agricultural industry toward more sustainable, tech-driven  
solutions, such as precision agriculture, which uses sensors, drones, and machine learning models to detect and  
Page 2098  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
manage pests more effectively. This shift highlights the need for continued innovation in pest control to maintain  
crop health and ensure long-term agricultural sustainability (Karar et al., 2022).  
Pest detection is a critical component of integrated pest management (IPM) in agriculture, as it helps identify  
infestations early and ensures timely intervention to protect crops. Various methods are employed for detecting  
pests, ranging from traditional manual techniques to advanced technological approaches (Anwar and Masood,  
2023). Each method has its advantages and limitations, and modern agriculture increasingly incorporates  
automation and technology to improve accuracy and efficiency. Machine learning and computer vision are  
modern methods used to detect pests through image recognition and data analysis. Pre-trained neural networks  
are used to analyze images of crops, identifying pests or pest damage with high accuracy (Bhoi et al., 2021;  
Sochima et al., 2025; Deepika and Arthi, 2022).  
Deep learning is increasingly being integrated into autonomous robotic systems for pest management (Kekong  
et al., 2019). Robots equipped with cameras and deep learning algorithms can navigate fields, scanning crops  
for signs of pest infestations (Sun et al., 2023). These systems are not only capable of detecting pests but also  
performing targeted pest control measures, such as spraying pesticides or releasing beneficial organisms. This  
approach enhances the efficiency of pest management and reduces labor costs, while minimizing the  
environmental impact by focusing pest control efforts only where necessary.  
Pest has remained a major issue for farmers over the years and has continued to witness increased research  
attention with diverse proposals to help address the problem. Among the most recent studies reviewed, Kumar  
et al. (2023) applied light weight YOLOV-5 for the classification of pest using field adaption method. While  
this work reported good classification success for pest detection in a farm, there is gap due to false alarm, region  
specification of the model that is, the model is limited to best within the region of data collection),  
misclassification, and may not be reliable for real time remote sensing applications.  
RESEARCH METHOD  
The system will be made of three major components which are the classification model, real time notification  
system and UAV. The proposed classification and notification model will use image data of pest collected from  
the Federal College of Agriculture, Ishiagu, Ebonyi State, Nigeria and then fine tune existing pest data which  
will be collected in roboflow repository to develop a comprehensive data model. The data will be used to train a  
deep learning algorithm, specifically YOLO-V10 which is a more recent and advance version of YOLO-V, then  
it will be trained to generate model for the real time classification of pest in the farm. To make the model reliable,  
a pest detection decision algorithm will be developed, using the classification model as a foundation to decide if  
the farm is infected with disease or not. In the next phase of the proposed system, IoT algorithm will be developed  
and integrated with the classification model to facilitate real-time notification to the farmer on the event of pest  
in the farm. Finally, a decision-based model will be developed which inform the farmer on the right pesticide to  
be applied on the farm to help control the pest. The proposed system block diagram was presented in Figure 1.  
Figure 1: Block diagram of the proposed system  
Page 2099  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Figure 1 presents the block diagram of the proposed system. This system began with data collection of primary  
and secondary dataset. Both datasets will be processed through annotation and labelling, then collectively  
integrated to create a new data model YOLOv-10which will be trained with the data to generate model for real  
time pest classification. To address issues of false alarm, a decision-based algorithm will be developed and  
integrated with the deep learning-based classifier. This will facilitate accurate detection of pest in the farm and  
then address issues of reliability as characterized in the existing system. To ensure real time notification of the  
farmer on the issue of pest, an IoT algorithm will be developed which used email to notify the farmer of the  
issues of pest. Finally, the model will be deployed into the drone for remote sensing purposes.  
Figure 2: Proposed system for the management of pest in agriculture  
Figure 2 presents the proposed system for the management of pest in agriculture using dataset of pests, deep  
transfer learning, an IoT algorithm and control measures in this order. The comprehensive data model which  
composed of localized pest in Nigeria was used to train a YOLOV-10 algorithm and generate a model for the  
real time classification of pest in the farm. The model was used as foundation to develop a decision-based model  
for the classification of pests in the farm, while the classified output was communicated to the farmer using IoT  
algorithm, which will be incorporated with the UV for real time monitoring purposes.  
The Research Area  
The research was conducted at the Federal College ofAgriculture, Isiagu, located in Ivo Local Government Area  
of Ebonyi State, Nigeria. This institution is known for its agricultural research and training, making it an ideal  
site for collecting pest-related image data in real field conditions. The geographical coordinates of the location  
are approximately latitude 6.0055° N and longitude 7.5609° E. The region falls within the tropical rainforest  
zone of southeastern Nigeria, characterized by high humidity and substantial rainfall, which creates a suitable  
environment for both crop cultivation and pest proliferation. These conditions provided a realistic setting for  
observing and capturing diverse pest infestations affecting various crops. Figure 3 presents the research area  
Map.  
Figure 3: Research Area Map  
Page 2100  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Data collection  
The data used for this work are secondary and primary data. The primary data was collected at the Federal  
College ofAgriculture, Ishiagu, Ebonyi State, Nigeria. The sample size of data collected was 209 images of pest  
(aphids, armyworms, bees, beetle, bugs, lopers, caterpillar, citrus canker, beetles, corn earth-worms and corn  
borers). The experimental setup for this work constitutes several materials such as a universal serial bus, high  
definition camera, Raspberry Pi, the farm, power supply bank, and laptop. The secondary data is the insect pest  
detection dataset from Kaggle repository (source: https://www.kaggle.com/datasets/cubeai/insect-pest-  
detection-for-yolo). The sample size of the data is 17641. The overall data size is 17850. The data sample is  
presented as Figure 4;  
Figure 4: Pest data from secondary source (Source: Kaggle)  
Data Preparation  
The data collected was prepared through a systematic process of annotation and labeling using the Roboflow  
Toolbox. This tool enabled precise identification and classification of various pests in the images, such as aphids,  
armyworms, bees, beetles, bugs, loopers, caterpillars, citrus canker, corn earworms, and corn borers. Each image  
was annotated with bounding boxes to indicate the exact location of pests, and labeled accordingly to reflect the  
pest type. After annotation, the dataset was split into training, validation, and testing sets to ensure efficient  
machine learning model development. Image augmentation techniques such as rotation, flipping, and brightness  
adjustment were also applied within Roboflow to increase dataset diversity and improve model generalization  
during training.  
The Computer Vision Systemfor Real-Time Pest Detection and Classification using Pre-Trained  
Technique  
This section presents the computer vision algorithm model of this work using pre-trained model. The type of  
pre-trained model adopted is YOLOV-10 (Ebere et al., 2025). The architecture is presented in Figure 5.  
Figure 5: Architecture of the YOLOV-10 (Ebere et al., 2025)  
Page 2101  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Figure 5 illustrates the architecture of the YOLOv10 model, which is organized into four key components: the  
input, backbone, neck, and head. Each of these plays a distinct role in the object detection pipeline, contributing  
to the model’s overall performance and accuracy. The input stage defines the dimensions of the image provided  
to the model. Typically, images are resized to a fixed resolution (640 × 640 pixels) to ensure consistency during  
training and inference. This input is then passed into the backbone for further processing.  
The backbone is responsible for feature extraction. It processes the input image using a series of convolutional  
layers to capture low-level and mid-level features such as edges, textures, and object shapes. This stage plays a  
crucial role in building a detailed representation of the image content. Following the backbone is the neck, which  
performs multi-scale feature fusion. It combines features from various layers of the backbone to ensure that the  
model can detect both small and large objects effectively. This aggregation improves the model’s robustness  
when working with objects of varying sizes and spatial locations. The head produces the final detection output.  
It interprets the processed features to generate bounding boxes, class probabilities, and confidence scores. These  
outputs are then used to identify the presence and position of objects in the image.  
Several specialized modules support the internal functioning of the model. The Conv layers (convolutional  
operations) are used for scanning the input and extracting essential features. The C2f module (Cross-Stage Partial  
with Fusion) houses the bottleneck layers, which often incorporate pre-trained models to improve learning  
efficiency and reduce overfitting. The Concat operation merges features from different layers, enriching the  
feature map with more diverse information. Finally, the SPPF (Spatial Pyramid Pooling Fast) module is a critical  
component at the end of the backbone, designed to extract features across multiple receptive fields by applying  
pooling at various scales. This enables the model to better understand spatial hierarchies within the image.  
Pest Management Model  
The pest management model was developed through expert consultation at the Research area (FCA, Ishiagu).  
Qualitative data was collected on pest and their recommended control solution. The results of the data collected  
were reported in the Table 1.  
Table 1: Pest control Data (Source: FCA, ISHIAGU)  
S/N  
1
Pest  
Recommendation pesticides  
Insecticidal soap  
Ampligo  
Aphids  
2
Armyworms  
bees  
3
No control method for now  
CypeForce  
4
Canker  
5
Bugs  
Imidacloprid  
6
Lopers  
CaterpillarForce  
CaterpillarForce  
CypeForce  
7
Caterpillar  
Citrus canker  
Beetles  
8
9
Cypermethrin DudUALL 450 EC  
No chemical control  
CypeForce  
10  
11  
Corn earth worms  
Corn borers  
The rule-based recommendation algorithm is presented as;  
Page 2102  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Rule-Based Recommendation Algorithm  
1. Initialize the system.  
2. Connect to pest classification models.  
3. Load Table 1.  
4. Column 1: Serial Number (S/N)  
5. Column 2: Pest  
6. Column 3: Recommended Control Pesticide(s)  
7. For each pest in Table 1:  
8. Identify the pest.  
9. Apply the corresponding pesticide(s) listed in Column 3.  
10. End For  
11. Stop the process.  
UAV for Remote Sensing and Pest Monitoring  
The modelling of the UAV was adopted from Osisiogu et al. (2019) and developed using structural method via  
architectural universal modelling diagram to present the various section of the system as shown in the Figure 6;  
Figure 6: Architectural Model of the Unmanned Vehicle (Osisiogu et al., 2019)  
The UV model in Figure 6 is made of the various sections which are the power supply, control system, actuator  
system, sensors and software defined radio. The power supply is a mini drone battery which supplied regulated  
direct current to the system. The control section of it was made of pitch and yaw adjustment controller and also  
a fuzzy logic control system (Habor et al., 2021). The pitch and yaw adjustment controller was used for the  
aerodynamic stability control of the whole system, while the fuzzy logic controller was used for the training and  
control of the sensor. The actuator mechanism is the kinematic section of the system which is composed of the  
propeller, DC motor, payload (Software Defined Radio (SDR)) and the landing gears. The SDR is a  
Page 2103  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
communication system equipped with 4G wireless devices which received command from the operators to  
control the drone and also sends the results of the data collected from the sensor for observation and analysis to  
the monitoring centre (Ulagwu- Echefu et al., 2021).  
Model of the UAV Behaviour in the Farm  
UAV is generally described with kinematics and dynamic model. The kinematic model established the geometric  
relationship between positions, velocity, orientation and angular rate without considering impact of external  
forces. It is dependent on the transformation of frames matrices to map out body velocities into inertia-frame  
position and relate Euler angles and angular rates changes with time. The dynamic model on the other hand is  
based on the Newton Euler laws model, the UAV translational and rotational acceleration to the total external  
forces and moments which acts on it. Collectively these formulations provide the foundation for the UAV  
modelling.  
The kinematic model of the UAV  
The kinematics model is developed considering the linear and angular kinematic motion model. The linear  
motion is represented considering velocity components (  
,
,
), along each axis within the coordinated frame  
of the body. The translational position ( , , ) for UAV is determined in an inertial reference frame. To relate  
these velocity vectors in the rate of change of position vector and body frame, rotational transformation and  
differentiation are applied as shown in Equation 1 (Ahmed et al., 2025).  
( )  
( ) =  
(1)  
Where x is the position along x axis, y is the position along y axis and z the position in the z axis.  
,
,
are  
the linear velocities of objects along the three axis ( , , ). is the rotational matrix transformation of the UAV  
body frame and represented in Equation 2 (Rahul et al., 2018; Gu et al., 2023).  
−
−
−
−
∅
∅
∅
∅
∅
∅
(
)
=
(2)  
∅
∅
∅
−
∅
∅
2.7.2. Angular motion of the UAV  
This describes the relationship between the Euler angles (∅, , ) from the three distinct coordinate’s frames and  
angular rates ( , , ) in the body frame and represented in Equation3.  
∅
( )  
=
( )  
(3)  
Where p is the roll rate and signifies the angular velocity within the x-axis, q is the pitch rate which is the angular  
velocity within y axis and corresponds to pitch axis. R is the yaw rate around the z axis angular velocity. ∅ is the  
roll angle, is the pitch angle and is the yaw angle. The transformation matrix of these frames to the body  
frame is represented as Equation 4;  
1
∅
∅
0
−
∅
= (  
)
(4)  
∅
∅
0
Page 2104  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Dynamic model  
This section presents the dynamic model of the UAV, explaining how the system behaves under external force,  
based on the Newton-Euler Law of motion. The model was adopted from Ahmed et al. (2025) and developed  
with assumption that the ground level is flat. The movement of the UAV therefore is converted from body frame  
to earth frame through rotational matrices. Based on these relations, the dynamic movement of the UAV is  
defined Equation 5  
d
=
+
∗
(5)  
/
dt  
b
Where P is the frame derivatives as it changes, while w is the angular velocity in the inertia frame. To model the  
UV motion during translation, the second Newton’s law was applied as shown in Equation 6;  
∑
(
)
=
=
(6)  
Where F is the external force which arising from aerodynamic forces, gravity and propulsion control of the UAV.  
m is mass, is inertia velocity derived from Equation 5 and expressed in angular velocity defined in Equation  
7, while the body frame is defined in Equation 8.  
=
=
+
∗
(7)  
/
∑
+
∗
(8)  
/
(
)
,
(
)
,
(
)
Where  
,
,
,
,
, ,  
the velocity vector in the body frame. Consequently,  
/
the three equations that govern translational dynamics can be outlined as follows;  
̇
−
−
−
1 ( )  
̇
( )  
(
)
=
+
(9)  
̇
The rotational motion model was also developed with Newton second law which postulated Equation 10.  
(
)
=
=
(10)  
Where M is the external moment. This equation is expanded applying Equation 10 to the rate of change of  
angular momentum with respect to time in the inertia frame as shown in Equation 11;  
=
+
∗
(11)  
/
Where  
al., 2020).  
is the angular momentum as a result of the inertia matrix (J) by the angular vector (Lopez-Briones et  
(12)  
=
/
From Equation (11 and12), he Equation 13 and 14 are obtained as;  
/
∑
=
+
∗
(13)  
(14)  
/
/
̇
= ( ) = −1[−  
* (  
) +  
]
/
/
̇
Page 2105  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
(
)
Where  
=
(15)  
The J diagonal matrix represents moments of inertia, which indicates the UAV resilience to acceleration within  
specified axes of motion (Chi et al., 2015). The off-diagonal matrix is the inertia product with certain inertia  
neglected with assumed zero minimal impact.  
0
0
0
0
0
0
(
)
Where J =  
(16)  
Develop a real time notification algorithm  
In smart agricultural systems and other intelligent monitoring platforms, timely communication of critical events  
is essential to enable users to take prompt and effective action. This section presents a real-time notification  
algorithm designed to automatically alert users through email when predefined events are detected by the system.  
The algorithm ensures that relevant stakeholders are informed with actionable recommendations, helping them  
respond efficiently to emerging situations. The notification mechanism operates by continuously monitoring  
system conditions, generating personalized alert messages, and dispatching them through secure email channels.  
Below is the stepwise process of the notification algorithm. The stepwise is presented as;  
Real-Time Email Notification Algorithm (Stepwise)  
1. Trigger Event Detection%% Monitor for specific trigger conditions  
2. IF system_event == TRUE AND confidence_level ≥ THRESHOLD  
3. → proceed to notification  
4. Set user information  
5. Retrieve User Information %% Fetch user data (e.g., email address, name)  
6. Generate Message Content %% Generate the email message content  
7. Send Email Notification %% Connect to an SMTP server and send the email  
8. Log Notification Record%% Store a log entry for traceability and confirmation.  
9. End  
System Integration  
This section presents the integrated system developed for real-time pest monitoring and decision support. The  
integration process began with the successful training and validation of the pest detection model using the  
annotated pest dataset. Once the model achieved satisfactory performance, it was exported and deployed on a  
lightweight edge device compatible with the UAV system. The email notification module was then integrated by  
linking it to the model's inference output. Whenever pests were detected, the system automatically generated an  
alert and sent an email containing actionable information to the farmer's registered address. This was achieved  
using a Python-based SMTP script embedded within the inference pipeline. Figure 7 presents the architecture of  
the UAV.  
Page 2106  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Figure 7: Architecture of the UAV with pest detection and notification module  
Training of the model with the pest dataset  
The YOLOv10 model was trained using a custom pest dataset consisting of annotated images representing  
various pest classes such as aphids, armyworms, beetles, bugs, and bees. The dataset was pre-processed to ensure  
uniformity in image dimensions, quality, and labelling format. All images were resized to 640 × 640 pixels to  
align with the model’s input requirements, while data augmentation techniques such as horizontal flipping,  
scaling, and rotation were applied to enhance the model’s generalization capability. The dataset was divided into  
training, validation, and test sets in a ratio of 70:20:10. This ensured that the model was trained on a substantial  
portion of the data while being evaluated on unseen samples for performance assessment. Labelling was  
performed using the Roboflow annotation tool, and the dataset was converted to the YOLO format, which  
includes text files containing class indices and normalized bounding box coordinates for each image.  
Training was conducted using the PyTorch framework on a GPU-enabled environment to accelerate  
computation. Abatch size of 16 and an initial learning rate of 0.001 were used. The Adam optimizer was selected  
for efficient convergence, and the learning rate was dynamically adjusted using a cosine annealing scheduler.  
The model was trained over 100 epochs, with performance metrics such as training loss, validation loss, mean  
Average Precision (mAP), and F1-score monitored at each epoch to guide early stopping and model  
checkpointing.  
Result Of The Yolov-10 Training  
This section discusses the performance results of the YOLOv10 benchmark model after training it on the  
annotated pest dataset. The training process was carefully monitored, and several evaluation metrics were  
recorded to assess the model’s accuracy in detecting and classifying different pest species. The results  
demonstrate how well the model learned from the data and its capacity to generalize to new, unseen images.  
During the training phase, the model achieved a bounding box loss of 0.65. This metric indicates the accuracy  
with which the model could predict the location of pests in an image. A low bounding box loss value signifies  
that the model was effective at learning object localization. Similarly, the training classification loss was recorded  
at 0.64, showing the model’s ability to correctly assign labels to the pests it identified. The training focus loss,  
which evaluates how well the model concentrated on the most relevant parts of the image, stood at 0.54,  
reflecting good spatial awareness during training. Figure 8 presents the training results.  
Page 2107  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Figure 8: Training result of the benchmark YOLOV-10N  
On the validation dataset, the model recorded a bounding box loss of 1.002. While slightly higher than the  
training value, this is still within an acceptable range and suggests that the model has not significantly overfitted  
the training data. The validation classification loss was 0.68, and the focus loss increased to 1.2. These validation  
losses indicate that the model maintained reasonable accuracy in pest identification on new data, although the  
increased focus loss suggests some difficulty in attending to exact object regions during validation.  
In terms of detection accuracy, the model achieved a precision score of 0.84. This means that 84% of the objects  
detected by the model were actual pests, with few false positives. The recall was 0.82, indicating that the model  
was able to successfully detect 82% of all true pest instances in the dataset, minimizing false negatives.  
Furthermore, the model scored 0.83 on mean Average Precision at an intersection over union threshold of 0.5  
(mAP@50). This shows that the model was highly effective in identifying pests with at least 50% overlap with  
the ground truth boxes. When evaluated across a more rigorous range of IoU thresholds (mAP@50–90), the  
model recorded a score of 0.60, which still reflects solid performance across various levels of detection difficulty.  
The YOLOv10 benchmark model performed strongly across all major training and validation metrics. The losses  
remained within acceptable ranges, and the evaluation scores confirm that the model is capable of accurately  
identifying and classifying pests in real-time. These results support the model’s integration into a practical  
decision support system for pest management using UAVs or smart agricultural surveillance platforms.  
The UAV simulation results  
This section presents the result of the UVAapplied for remote sensing of the farm environment for pest detection.  
The results were generated with Simulink as the tool and the simulation parameters used was adopted from Oti-  
Owom et al. (2024) and was applied to simulate the flying of the UAV, while capturing data from the camera  
and then classifying pest. Figure 9 present the Simulink block diagram of the UAV, while Table 2 presents the  
simulation parameters.  
Page 2108  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Figure 9: Simulink block diagram  
Figure 10: Result of the UV in Simulink  
The Figure 10 presents the results of UAV in Simulink. This result was achieved from the simulation using  
simulation parameters of UV adopted from Osisiogwu et al. (2019) to run the program in Simulink 3D. The  
parameters used to run the simulation program are all reported in Table 2.  
Table 2: Simulation parameters (Osisiogwu et al., 2019)  
Parameter  
Value / Description  
DJI Phantom 4 Pro  
20 – 30  
Unit / Type  
UAV Model  
Flight Altitude  
Camera Resolution  
Camera Type  
meters  
3840 × 2160  
pixels (4K UHD)  
RGB with fixed focal length  
Page 2109  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
Frame Rate  
30  
frames per second (fps)  
Image Capture Interval  
Flight Speed  
2
seconds  
m/s  
3 – 5  
Flight Time per Session  
Battery Capacity  
20 – 25  
minutes  
mAh  
5870  
GPS Accuracy  
±1  
meter  
hectares  
images  
-
Area Coverage per Flight  
Number of Images Captured  
Detection Algorithm  
2 – 3  
209  
YOLOv10n + Improved C2F  
The UAV-based pest monitoring system proposed using YOLOv10 was effective in the real-time detection with  
the precision and recall values of 0.84 and 0.82, respectively. The combination of IoT-based alerts and rule-based  
pesticide suggestions offers the entire decision-support system to the farmers. The UAV simulation established  
the stable flight behaviour and dependable image data collection to monitor the pests in remote areas which  
confirmed that the system is ready to be utilised in realistic setting of precision agriculture.  
CONCLUSION  
This paper came up with a smart and technology based real-time pest detection and management in agriculture  
through deep learning, IoT and integration of UAVs. The study focused on the weaknesses of conventional pest  
management practises and models regionally by presenting an extensive framework of integrating image-based  
pest detection and automated decision-making and notification systems. Primary sources were used to form a  
hybrid dataset at the Federal College of Agriculture, Ishiagu, and secondary data at the Kaggle repository. The  
data were labelled and trained on the YOLOv10 model, the type of which was selected due to its high speed and  
its ability to perform the task of object detection.  
The trained YOLOv10 model had good performance scores of 0.84, 0.82, and 0.83 of precision, recall, and  
mAP@50 respectively, which proved to identify and classify different pest species accurately. A  
recommendation algorithm based on rules was also designed to recommend appropriate actions to be taken to  
deal with each of the identified pests. To guarantee the farmers responded in time, an email notification algorithm  
with the help of IoT was incorporated with a detection system, allowing real-time notification of the farmers in  
case pests were detected. The simulated UAV model based on Simulink gave a dependable aerial pest surveying  
model with steady flight controls and onboard image processing.  
To summarise, the suggested system is able to combine innovative computer vision, IoT, and UAV to improve  
efficiency in pest monitoring and management in agriculture. The capabilities of the model to identify pests with  
high precision and provide useful insights that can be acted on by farmers in real-time can be used to show its  
feasibility in the application of the model to address precision farming. In addition to minimising the need of  
chemical pesticides, it reduces losses and enhances informed decision-making to support sustainable agriculture  
by manipulating aspects of crop health and productivity.  
REFERENCES  
1. Ahmed, E., Ehab, S., Mohamed, R., & Ahmed, M. (2025). Model-based simulation and validation of  
small fixed wing UAV. Engineering Science and Military Technologies, 9(1).  
https://doi.org/10.21608/EJMTC.2024.243244.1267  
Page 2110  
www.rsisinternational.org  
INTERNATIONAL JOURNAL OF RESEARCH AND INNOVATION IN APPLIED SCIENCE (IJRIAS)  
ISSN No. 2454-6194 | DOI: 10.51584/IJRIAS |Volume X Issue X October 2025  
2. Alanazi, A., Shakeabubakor, A., Abdel-Khalek, S., & Alkhalaf, S. (2023). IoT enhanced metaheuristics  
with deep transfer learning based robust crop pest recognition and classification. Alexandria Engineering  
Journal, 84, 100–111. https://doi.org/10.1016/j.aej.2023.11.008  
3. Anwar, Z., & Masood, S. (2023). Exploring deep ensemble model for insect and pest detection from  
images. Procedia Computer Science, 218, 2328–2337.  
4. Bhoi, S., Jena, K., Panda, S., Long, H., Kumar, R., Subbulakshmi, P., & Jebreen, H. (2021). An Internet  
of Things assisted unmanned aerial vehicle-based artificial intelligence model for rice pest detection.  
Microprocessors and Microsystems, 80, 103607. https://doi.org/10.1016/j.micpro.2020.103607  
5. Chi, S., Lee, S., Ryu, H., Shim, H., & Ha, C. (2015). Dynamics and simulation of the effects of wind on  
UAVs and airborne wind measurement. The Japan Society for Aeronautical and Space Sciences, 58(4),  
187–192.  
6. Deepika, P., & Arthi, B. (2022). Prediction of plant pest detection using improved mask FRCNN in cloud  
environment. Measurement: Sensors, 24, 100549. https://doi.org/10.1016/j.measen.2022.100549  
7. Ebere Uzoka, C., Anoliefo, E., Udanor, C., Chijindu, T., & Nwobodo, L. (2025). Ablind navigation guide  
model for obstacle avoidance using distance vision estimation based YOLO-V8n. Journal of the Nigerian  
Society of Physical Sciences, 2292–229. https://doi.org/10.46481/jnsps.2025.2292  
8. Gu, Y., Zhang, G., Bi, Y., Meng, W., Ma, X., & Ni, W. (2023). Pitch mathematical modeling and dynamic  
analysis of a HALE UAV with moving mass control technology. Aerospace, 10(11), 918.  
https://doi.org/10.3390/aerospace10110918  
9. Harbor M.C, Eneh I.I. Ebere U.C. (2021). Precision control of autonomous vehicle under slip using ANN.  
International Journal of Research and Innovation in Applied Science (IJRIAS). Vol 6; Issue 9.  
https://rsisinternational.org/journals/ijrias/DigitalLibrary/volume-6-issue-9/62-68.pdf  
10. Karar, M., Abdel-Aty, A., Algarni, F., Hassan, M., Abdou, M., & Reyad, O. (2022). Smart IoT-based  
system for detecting RPW larvae in date palms using mixed depth wise convolutional networks.  
Alexandria Engineering Journal, 61, 5309–5319. https://doi.org/10.1016/j.aej.2021.10.050  
11. Kekong P.E, Ajah I.A., Ebere U.C. (2019). Real-time drowsy driver monitoring and detection system  
using deep learning based behavioural approach. International Journal of Computer Sciences and  
Engineering 9 (1), 11-21; http://www.ijcseonline.isroset.org/pub_paper/2-IJCSE-08441-18.pdf  
12. López-Briones, F., Sánchez-Rivera, M., & Arias-Montano, M. (2020). Aerodynamic analysis for the  
mathematical model of a dual-system UAV. 17th International Conference on Electrical Engineering,  
Computing  
Science  
and  
Automatic  
Control  
(CCE),  
1–6.  
https://doi.org/10.1109/CCE50788.2020.9299157  
13. Oti-Owom, J. O., Eke, J., & Umeozulu, A. (2024). Modelling of intelligent robot for gas pipeline leakage  
detection and control using deep learning technique. Journal of Engineering and Technology, 1(7).  
14. Rahul, A., Divya, P., & Shishirkumar, K. (2018). Mathematical modeling and simulation of quadcopter-  
UAV using PID controller. 4th International Conference on Engineering Confluence & Inauguration of  
Lotfi Zadeh Center of Excellence in Health Science and Technology (LZCODE) – EQUINOX 2018.  
15. Sochima V.E. Asogwa T.C., Lois O.N. Onuigbo C.M., Frank E.O., Ozor G.O., Ebere U.C. (2025)”;  
Comparing multi-control algorithms for complex nonlinear system: An embedded programmable logic  
control  
applications;  
DOI: http://doi.org/10.11591/ijpeds.v16.i1.pp212-224  
16. Sun, L., Cai, Z., Liang, K., Wang, Y., Zeng, W., & Yan, X. (2023). An intelligent system for high-density  
small target pest identification and infestation level determination based on an improved YOLOv5 model.  
Expert Systems With Applications, 239, 122190. https://doi.org/10.1016/j.eswa.2023.122190  
17. Ulagwu-Echefu A., Eneh .I.I. Ebere U.C. (2021). Enhancing realtime supervision and control of  
industrial processes over wireless network architecture using model predictive controller. International  
Journal of Research and Innovation in Applied Science (IJRIAS); vol 6; Issue 9.  
https://rsisinternational.org/journals/ijrias/DigitalLibrary/volume-6-issue-9/56-61.pdf  
Page 2111  
www.rsisinternational.org